AttnLRP: Attention-Aware Layer-wise Relevance Propagation for Transformers
Our new method, AttnLRP, is the first to faithfully and holistically attribute not only input but also latent representations of transformer models with the computational efficiency similar to a single backward pass. We demonstrate that our...
Reactive Model Correction: Mitigating Harm to Task-Relevant Features via Conditional Bias Suppression
DNNs are prone to relying on spurious correlations in data, posing risks in critical applications. Post-hoc methods exist to mitigate this without retraining but can globally shift latent features distributions, harming model performance. We...
Explainable Concept Mappings of MRI: Revealing the Mechanisms Underlying Deep Learning-Based Brain Disease Classification
While recent studies show high accuracy in the classification of Alzheimer's disease using deep neural networks, the underlying learned concepts have not been investigated. We separated Alzheimer's patients (n=117) from normal controls (n=219) by...
PURE: Turning Polysemantic Neurons Into Pure Features by Identifying Relevant Circuits
Neurons in deep neural networks can act polysemantically, meaning that they encode for multiple (unrelated) features. As such, understanding the inner workings of machine learning models becomes more difficult. We present PURE to turn...
Model guidance via explanations turns image classifiers into segmentation models
Heatmaps generated on inputs of image classification networks via explainable AI methods have been observed to resemble segmentations of input images in many cases. We apply the "Right for the Right Reason" paradigm of imposing additional losses...
Unlocking the Potential of Local CSI in Cell-Free Networks with Channel Aging and Fronthaul Delays
Centralized or distributed precoding? This is perhaps the most heated debate in the Cell-free Massive MIMO literature. However, in this work we argue that the best option may actually be a mix of the two. The reason is that centralized precoding,...
Gaussian Splatting Decoder for 3D-aware Generative Adversarial Networks
We present a novel approach that combines the high rendering quality of NeRF-based 3D-aware GANs with the flexibility and computational advantages of 3DGS. By training a decoder that maps implicit NeRF representations to explicit 3D Gaussian...
Towards Bridging the Gap between Near and Far-Field Characterizations of the Wireless Channel
Exploring near-field propagation is vital for 6G technologies like intelligent reflecting surfaces (IRS). Unlike far-field models, near-field models offer accuracy critical for applications such as beamforming and multiple-access, enhancing...
Optimized Detection with Analog Beamforming for Monostatic Integrated Sensing and Communication
This paper presents an optimization framework for analog beamforming in monostatic integrated sensing and communication (ISAC), which maximizes sensing detection under self-interference cancellation via superiorized projections onto convex...
Si3N4 Microring-Resonator-Based Integrated Photonic Sensor for Enhanced Label-Free Biofluid Analysis in the 850 nm Optical Band
We present an innovative integrated photonic sensor for point-of-care applications using silicon nitride (Si3N4) tailored for biofluids analysis in the 850nm optical band. With microring resonators (MRR) we detect smallest changes in the...